Point Cloud Segmentation using RGB Drone Imagery

Wudunn, Marc; Dunn, James; Zakhor, Avideh

doi:10.1109/icip40778.2020.9191266

Cited by 6 publications

(1 citation statement)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Work [15] proposes a multilayer perceptron for tower detection and classification using histograms of oriented gradients (HOG) to train two multi-layer perceptron neural networks: the first for background-foreground segmentation, and the second, for classifying within 4 different types of electric towers. Work [16] applies DL-based segmentation to RGB images used to create the point cloud itself, followed by back-projecting the pixel class in segmented images onto the 3D points. Work [17] uses a real-time DL scheme based on featured pyramid networks and YOLACT to detect electrical elements such as the cross-arms, insulators, transformers, and primary wires in close-range images.…”

Section: Related Workmentioning

confidence: 99%

Real-time LiDAR-based Semantic Classification for Powerline Inspection

Valseca

Paneque

Dios

et al. 2022

2022 International Conference on Unmanned Aircraft Systems (ICUAS)

View full text Add to dashboard Cite

Powerline inspection requires extracting accurate measurements of the distances between powerlines and between vegetation and powerlines and electrical towers. Existing automatic powerline inspection methods using manned helicopters, Vertical Take-Off and Landing (VTOL) vehicles, or quadrotors obtain these distances offline, days after LiDAR data gathering, using complex algorithms that prevent their online computation. This paper presents an efficient online processing scheme for unsupervised segmentation for powerline inspection using LiDAR-only data. It receives each point cloud from the LiDAR and outputs clusters of points classified into categories Powerlines, Towers, Vegetation, and Soil. Unlike existing approaches, our method relies on a combination of reflectivity and geometry, which simplifies object segmentation and enables online onboard execution. The method was experimented in sets of powerline inspection flights in environments with different conditions and vegetation. The proposed method succeeded in providing suitable online object segmentation involving 56% lower computational cost than existing learning-based methods.

show abstract

Section: Related Workmentioning

confidence: 99%